Evaluate JIT_MemCpy with wider copies using AVX2 on hardware that support it

[helloguo]

JIT_MemCpy on x64 Windows may benefit from using wider copies using YMM registers for certain copy lengths on hardware that support it. Requires CPU dispatch and may additionally require VZEROUPPER instruction to avoid transition penalties. dotnet/coreclr#7198

category:cq
theme:helpers
skill-level:expert
cost:medium

[nietras]

Can CPU dispatch not be done using the function pointer trick where on first call the CPU id is determined and this updates the function pointer, so following calls no longer do the CPU id check?

[TIHan]

Would be interesting to see, though the behavior should remain the same whether or not we want to handle overlapping buffers: #6701

[EgorBo]

Closed by #98623